Dan, This is a good start. There is going to be a wealth of data available - and it seems like you already have a framework for getting the relevant bits of data, which is the hardest part of projects.

From what I gather, the focus of the project is to find popular (or interesting?) videos by the same author. I would like some clarity as to what you for a classifier, and we can brainstorm how to fit this into a machine learning problem.

If you are going with popular - you could make a defined classifier saying if the video is popular or not - based on interactions, views, and comment count, etc. Interesting seems subjective to me, but perhaps it can be a metric like the above. You can have classes like controversial (e.g. equal likes and dislikes).

Perhaps sentiment of the title and first 50 comments could be used to predict if a video will be popular / interesting / controversial, etc?